SoftBank’s $4B DigitalBridge Acquisition to Supercharge AI Infrastructure

SoftBank Group has announced a definitive agreement to acquire digital infrastructure asset manager DigitalBridge Group for approximately $4 billion. This strategic acquisition is designed to significantly strengthen SoftBank’s position in the AI sector. By gaining control over critical digital assets like data centers, cell towers, and fiber networks, SoftBank aims to build the necessary infrastructure for training and deploying advanced AI models on a global scale, advancing its goal of achieving Artificial Super Intelligence (ASI). The transaction is expected to close in the second half of 2026, with DigitalBridge continuing to operate as a separately managed platform under its current CEO, Marc Ganzi.

Sources:

China Leads Humanoid Robot Race, Surpassing US in Commercial Deployment

China is establishing a significant lead in the commercial deployment of humanoid robots, while U.S. companies largely remain in pilot and testing phases. Chinese firm UBTECH Robotics has already shipped its 1,000th humanoid robot and is targeting production of 10,000 units by 2026. In contrast, prominent U.S. competitors like Tesla, Figure AI, and Agility Robotics are still conducting limited pilot programs. This gap was highlighted at a recent Silicon Valley robotics event where Chinese models were prominently featured, underscoring their market availability. This commercial head start enables Chinese companies to gather real-world data more rapidly, fueled by strong government backing that includes over $14 billion for AI and robotics initiatives. Companies like XPeng are also developing humanoid robots, signaling a broad strategic push into the sector.

Sources:

South Korean Self-Driving Startup RideFlux Secures ₩20 Billion for Commercialization

RideFlux, a South Korean autonomous driving software startup, has successfully raised 20 billion won in a pre-initial public offering (pre-IPO) equity investment round. The funding includes 10 billion won from existing shareholder Atinum Investment and 10 billion won from new investor Korea Development Bank. RideFlux stands out as the only company in Korea currently conducting unmanned, permit-based autonomous driving test operations with an empty driver’s seat. The company has an impeccable safety record, accumulating over 2,300 hours of autonomous driving data without any accidents. This new capital will help RideFlux complete its pre-IPO stage and supports its plan to go public on the KOSDAQ in the second half of 2026 as it prepares for large-scale commercialization of its self-driving technology.

Sources:

DevOps & Cloud News: AI on Kubernetes Certification and More

Key updates this week in DevOps, Kubernetes, and cloud infrastructure include:

  • The Cloud Native Computing Foundation (CNCF) announced a new certification for AI on Kubernetes, standardizing skills for running AI/ML workloads on the platform.
  • Palo Alto Networks and Google Cloud have expanded their strategic partnership to deliver integrated, AI-powered cloud security solutions.
  • A new report from Omdia reveals that global spending on cloud infrastructure services has hit $102.6 billion, a 25% increase in Q3 2025.

Sources:

Introducing Agent Sandbox: A New Open-Source Tool for Secure AI on Kubernetes

A new open-source Kubernetes controller, the Agent Sandbox, has been released to enable the secure deployment of AI agents. It provides a declarative API to manage isolated, stateful pods, making it ideal for executing untrusted, LLM-generated code securely. By leveraging technologies like gVisor, the Agent Sandbox creates a robust security barrier between the agent’s application and the cluster node’s operating system, mitigating potential security risks. Now a formal subproject of Kubernetes SIG Apps, it aims to standardize the management of stateful, isolated workloads.

Key features of the Agent Sandbox include:

  • Stable Identity: Provides consistent identities for stateful pods.
  • Persistent Storage: Ensures data is not lost across restarts.
  • Lifecycle Management: Includes creation, scheduled deletion, pausing, and resuming of agents.
  • Backend-Agnostic: Supports various isolation technologies like Kata Containers for flexible security and performance.

Sources:

Z.ai Open-Sources GLM-4.7 LLM for Production-Ready AI Development

Z.ai has released GLM-4.7, its latest open-source large language model (LLM) engineered for real-world software development and production workflows. This new version is optimized for practical engineering tasks, focusing on long-running task execution, stable tool calling, and multi-step reasoning. GLM-4.7 demonstrates significant improvements in code generation, complex reasoning, and agent execution stability. Benchmarks show its coding capabilities are comparable to leading models like Claude Sonnet 4.5. The model also introduces a “think before acting” mechanism to enhance its performance on complex tasks within coding frameworks, making it a more practical and dependable AI tool for developers and enterprises.

Sources: